purely curated selection, based on editorial policy. While the latter is evidently more sub-
jective, it usually involves real journalism—that is, journalists becoming well informed
about particular topics. An interesting question regards the benefits of these two forms
of news selection for different classes of content. In particular, it is interesting to explore
whether both types can survive for a particular content category since users value
differentiation and/or multi-home, or whether one type tends to become dominant.
Regarding recommendations based on the specific user, a website is informed about
the history of the user’s browsing behavior. Therefore, it can tailor its recommendation
to this behavior and recommend stories or news according to the user’s past preferences.
In contrast to the case when other users recommend stories or news, here a user’s own
past behavior determines the stories that she becomes aware of. Again, this may lead to a
loss in plurality.
These devices to obtain information compete with search engines. As an example,
consider Amazon versus Google. Many users searching for books now directly use
Amazon’s website and no longer search on Google. In the case of media, a similar pattern
can be observed, with users who are looking for news bypassing the search engines and
immediately visiting the website of their preferred news provider. An interesting ques-
tion is how such behavior affects the bias of search engines and (potentially) of news
websites.
10.5. MEDIA PLATFORMS MATCHING ADVERTISING TO CONTENT
The success of a firm’s advertising campaign is driven mainly by the effectiveness of its
ads. Foremost, the recipients of the ads (i.e., the potential consumers) should be primarily
individuals or companies with an inherent interest in the firm’s products. Otherwise,
informing potential consumers about characteristics of the product is unlikely to lead
to actual purchases. To increase advertising effectiveness and reduce wasted impressions,
firms match their advertisements to content on media outlets in such a way that con-
sumers who are interested in the content are also likely to be interested in the advertised
product. This practice is called content matching or tailoring and can be seen as a particular
instance of targeting.
Consider, for example, a local bookstore. The store has higher returns from placing
ads in a local newspaper than in a global one. The local newspaper is read by a local audi-
ence, which consists of the potential consumer group for the bookstore. By contrast, a
large portion of ads in the global newspaper are wasted since many readers do not live in
the vicinity of the bookstore. Similar examples apply to content instead of geography.
The advertisements of a cosmetics company are usually more effective in a women’s mag-
azine than in a computing magazine, and a sports apparel manufacturer’s ads are likely to
be more effective during televised sports than during a comedy show. However, in tra-
ditional media, the degree and effectiveness of such tailoring is limited. As argued by
494 Handbook of Media Economics
Goldfarb (2014), for example, the distinguishing feature of Internet advertising is its
reduction in targeting costs compared to traditional media.
On the Internet, targeting is not limited to linking advertising to specific content.
Advertisements can be targeted to the intentions of the consumer (reader/viewer/
listener) as inferred from past behavior or based on specific circumstances, such as the
weather conditions at the consumer’s location. For example, media platforms can expose
different users to different advertisements, even when those users browse the same web-
site at the same time. The particular advertisement can be conditioned on many different
parameters. For instance, the website may engage in geo-targeting and display advertise-
ments relevant to the user’s geography (inferred from IP addresses). Similarly, the website
may keep track of ad exposure to users, thereby reducing repetitious exposure of ads, or
search engines may display ads conditional on queries conducted, a practice called key-
word advertising. Clearly, both practices lower the number of wasted impressions, allow-
ing the website or search engine to charge higher prices to advertisers, everything else
given. A highly debated form of targeting is called behavioral targeting. Here, a website
customizes the display advertisement to information collected in the past about a user.
The website uses cookies based on pages that the user has visited and displays ads that
could be of particular interest to the user;
45
cookies are small pieces of data sent from
a website, which track the user’s activities. These cookies give precise information about
the user’s past web-browsing behavior and therefore about her preferences. We analyze
implications of behavioral targeting in
Section 10.6.
In
Section 10.5.1, we discuss different formalizations of targeting (in the wider sense),
focusing on tailoring on the Internet and how it differs from general tailoring. In
Section 10.5.2, we then discuss the practice of “keyword targeting” in more detail.
10.5.1 How to Formalize Targeting
In the economics literature, targeted advertising has been shown to be able to segment the
market.
Esteban et al. (2001) consider targeted advertising by a monopolist and show that
the monopolist will target primarily consumers with high reservation values, thereby
extracting a higher surplus.
Targeting also affects market outcomes under imperfect competition between adver-
tisers. In particular, segmentation due to targeting may relax product market competition
and thus allow firms to charge higher equilibrium prices.
Iyer et al. (2005) consider a
model with two competing firms that need to advertise to inform consumers about
the existence of their products. There are three different consumer segments. Consumers
belonging to the first have a high preference for the first firm in that its members consider
buying only from that firm; those belonging to the second have a high preference for the
second firm; and those belonging to the third are indifferent between the firms, and buy
45
For an in-depth discussion and analysis of behavioral targeting, see Chen and Staellert (2014).
495
The Economics of Internet Media
the lower-priced product.
46
Advertising is costly to firms. Iyer et al. (2005) show that
without targeting, equilibrium profits are zero because firms spend their entire product
market profit to inform consumers. By contrast, with the possibility of targeting con-
sumers, firms advertise with a higher probability to the market segment that prefers
the firm’s product than to the indifferent consumers, enabling the firms to reap strictly
positive profits.
Roy (2000) and Galeotti and Moraga-Gonzalez (2008), analyzing
different models, also show that targeting can lead to full or partial market segmentation,
allowing firms to obtain positive profits.
47
In what follows, we provide a more detailed discussion of the models by Athey and
Gans (2010)
and Bergemann and Bonatti (2011); both works explicitly consider targeting
strategies on the Internet. The former focuses on the supply side and keeps consumer
demand simple, whereas the latter explicitly models the demand side and keeps the supply
side simple.
48
We then briefly discuss the model by Rutt (2012).
Athey and Gans (2010) present a model that is cast in terms of geo-targeting. How-
ever, it can be adjusted to other forms of targeting. Specifically, consider a set of localities
x 2 1, …, X
fg
, where each locality consists of N consumers. In each locality, there is one
local media outlet. There is also one general outlet denoted by g, which is active in all
locations. Consumers single-home—that is, they visit only one outlet. The market shares
for local and global outlets in each locality are the same and given by n
g
for the global
outlet and Nn
g
for the local outlet.
Each advertiser i is only local and therefore values only impressions to consumers in
the respective locality. The value to advertiser i of informing a consumer is v
i
. Outlets
track advertisers, which implies that they offer each advertiser a single impression per
consumer. There is a continuum of advertisers with values v
i
2 0, 1½with cumulative
distribution function F(v
i
). Each outlet chooses the number of ads, a
j
, j 2 1, …, X, gfg,
that can be impressed on a consumer. We denote by p
j
the impression price of outlet
j. Finally, the probability of informing a consumer with an impression on outlet j is given
by ϕ
j
. There is no nuisance of advertising.
For each local outlet l, the probability of informing a consumer equals 1. Instead, for
the global outlet, this probability depends on targeting being possible or not. If targeting is
46
Chen et al. (2001) consider a similar demand structure to analyze the implications of targeting. In contrast
to
Iyer et al. (2005), they assume that firms can charge different prices to the consumers in different seg-
ments and show that imperfect targeting softens competition.
47
Other models of targeted advertising include van Zandt (2004), who analyzes information overload;
Gal-Or and Gal-Or (2005), who analyze targeting by a common marketing agency; and Johnson
(2013)
, who studies ad-avoidance behavior by consumers when targeting is possible. The latter will be
analyzed in
Section 10.6.1.
48
For an in-depth discussion of the different parameters influencing supply and demand of Internet adver-
tising, see
Evans (2009).
496
Handbook of Media Economics
not possible, this probability is ϕ
g
¼1=X because market shares are the same in all local-
ities. By contrast, if targeting is possible, the probability is ϕ
g
¼1.
Solving for the equilibrium number of ads when targeting is not possible, the
first observation is that an advertiser will buy impressions on outlet j if ϕ
j
v
i
p
j
. Since
ϕ
l
¼1 for local outlets, the total demand for impressions to a given consumer is 1 F(p
l
)
for a local outlet l. In equilibrium, demand equals supply, implying that 1 Fp
l
ðÞ¼a
l
or
p
l
¼F
1
1 a
l
ðÞfor a local outlet in a given locality. The profit function of outlet l is
therefore a
l
F
1
(1 a
l
), which is to be maximized over a
l
. We now turn to the global out-
let. An advertiser will buy impressions on outlet g if v
g
Xp
g
. Since a
g
are the impressions
per consumer and there are X localities, the overall demand is X(1 F(Xp
g
)), leading to an
equilibrium that is characterized by X 1 FXp
g
¼a
g
or p
g
¼ 1=XðÞ
F
1
1 1=XðÞa
g
. The profit function is a
g
(1/X)F
1
(1 (1/X)a
g
), which is to be
maximized over a
g
.
It is easy to see that the problem of the global outlet is the same as that of the local
outlets, adjusted by a scaling factor. Hence, a
g
¼Xa
l
and p
l
¼Xp
g
, implying that per-
consumer profits are the same. If targeting is possible, the problem of the global outlet
becomes exactly the same as that of the local outlet. In this case, the number of ads
and the ad price are the same for both types of outlets.
49
Athey and Gans (2010) obtain
that, without targeting, the global outlet expands its number of advertisements to X times
what a local outlet would provide due to wasteful impressions. Therefore, the price it
charges is only 1/X times that of a local outlet. However, per-consumer profits are
not affected, and the global outlet replicates the outcome of the local outlet.
To easily grasp the tradeoff in the model of
Athey and Gans (2010), it is instructive to
look at the model for the case in which advertising space is fixed for the global outlet. Is the
advertising price of the global outlet with targeting higher or lower than without? The
obvious effect is that advertising on the global outlet is less effective since advertisements
are mismatched with probability (X 1)/X. By contrast, advertising on a local outlet is
effective with probability 1. In addition to this efficiency effect, there is also a scarcity effect.
Without targeting, an advertiser from a locality competes with advertisers from other local-
ities for scarce advertising space. This increases the price on the global outlet. Formally,
comparing the advertising price on the global outlet with and without targeting gives
F
1
1 a
g
>
1
X
F
1
1
1
X
a
g
:
The advertising price with targeting is higher than that without targeting only if the last
inequality is satisfied. As shown by
Athey and Gans (2010), this holds true as long as a
g
is
not particularly high.
49
Note that the problems of local and global outlets are separated, implying that a change in the number of
ads of one outlet does not affect the number of ads on the other.
497
The Economics of Internet Media
Athey and Gans’s (2010) model demonstrates that targeting primarily allows an outlet
to reduce wasteful impressions. As long as there are no costs of these impressions, targeting
does not help an outlet to achieve higher profits. However, under many circumstances,
there are such costs. For example, in most of the models discussed above, consumers
dislike advertising. If there are nuisance costs to advertising, consumer demand is lower,
the larger the number of ads. Targeting then reduces this problem and allows the global
outlet to realize higher demand.
Athey and Gans (2010) provide other reasons for such
costs of impressions. Suppose, for example, that there is a constraint on advertising space
that prevents the global outlet from just raising its number of impressions. Targeting then
makes the use of the scarce advertising space more effective and allows the global outlet to
reap higher profits. (A similar reasoning holds for the case in which providing advertising
space is costly.) Alternatively, in the model presented above, demand across localities was
assumed to be homogeneous. However, a more realistic model would consider hetero-
geneous demand so that the global outlet has higher demand in some localities than in
others. This implies that advertisers in these localities have a higher willingness-to-pay
for advertising space. Thus, targeting allows the global outlet to price discriminate
between advertisers of different localities and obtain higher profits.
It is worth mentioning that targeting does not necessarily increase profits for the
global outlet. Consider an extension of the basic model in which outlets compete for
advertisers. This could be due to the fact that advertisers value, at most, one consumer
impression. As
Athey and Gans (2010) show, targeting can spur competition between
local and global outlets because the two types of outlets are vertically differentiated with-
out targeting. When implementing targeting, both outlets provide a similar service to
advertisers, leading to reduced prices. As a consequence, profits may fall with targeting.
Anecdotal evidence of excessively fine targeting reported by
Levin and Milgrom (2010)
supports the relevance of this result.
Athey and Gans’s (2010) model focuses on the supply side and reveals that increasing
the supply of advertising can be a substitute for targeting. Therefore, targeting is partic-
ularly effective if an outlet can increase its advertising space only by incurring a cost.
Bergemann and Bonatti (2011) pursue a different route by modeling the demand side in
a detailed way and keeping the supply side as simple as possible. In particular, they explicitly
introduce the idea that targeting on the Internet allows for unbundling of content, thereby
splitting a single advertising market into multiple ones. For example, readers of a traditional
newspaper have to buy the whole newspaper to access the content they are interested in.
Therefore, advertisers with niche products will probably find it too expensive to place an ad.
By contrast, online consumers may access (and pay for) only selected articles. This implies
that a producer of a niche product may find it profitable to pay for an ad that targets only the
consumer group interested in the particular article.
50
A similar effect holds for Internet TV.
50
This phenomenon has been called the “long tail of advertising”; see Anderson (2006). It also applies to
keyword advertising and behavioral targeting.
498
Handbook of Media Economics
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.